[
https://issues.apache.org/jira/browse/LUCENE-10171?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17503082#comment-17503082
]
ASF subversion and git services commented on LUCENE-10171:
----------------------------------------------------------
Commit 8afec33e747ec81c2301a4b099bd26b4195a556e in lucene's branch
refs/heads/main from Spyros Kapnissis
[ https://gitbox.apache.org/repos/asf?p=lucene.git;h=8afec33 ]
LUCENE-10171: OpenNLPOpsFactory should directly cache DictionaryLemmatizer
objects (#380)
Instead of caching dictionary strings and building multiple redundant
DictionaryLemmatizer objects.
Co-authored-by: Michael Gibney <[email protected]>
> Caching issue on dictionary-based OpenNLPLemmatizerFilterFactory
> ----------------------------------------------------------------
>
> Key: LUCENE-10171
> URL: https://issues.apache.org/jira/browse/LUCENE-10171
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/analysis
> Affects Versions: 9.0, 7.7.3, 8.10
> Reporter: Spyros Kapnissis
> Priority: Major
> Time Spent: 4h
> Remaining Estimate: 0h
>
> When providing a lemmas.txt dictionary file, OpenNLPLemmatizerFilterFactory
> caches internally only the string format of the dictionary, and not the
> DictionaryLemmatizer object. This results in parsing and creating a new
> DictionaryLemmatizer object every time the
> OpenNLPLemmatizerFilterFactory.create() is called.
> In our case, with a large lemmas.txt file (5MB) and the
> OpenNLPLemmatizerFilter used in many fields across our setup and in multiple
> collections (we use Solr), we had several random OOM issues and generally
> high server load due to GC activity. After heap dump analysis we noticed few
> thousands of DictionaryLemmatizer instances of around 80MB each.
> By switching the caching to the DictionaryLemmatizer instead of the String,
> we were able to resolve these issues. I will be attaching a PR for review,
> please let me know of any comments.
> Thanks!
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
